FPT.AI's Natural Language Processing performs best in English, Vietnamese
and Korean at the SHINRA2020-ML project. Photo courtesy of FPT. 

Vietnam's artificial intelligence platform FPT.AI ranked first at SHINRA2020-ML, a contest to classify 30 Wikipedia language entities in fine-grained categories, in Japan. 

The project is organized by Japan's RIKEN Center for Advanced Intelligence Project and Tohoku University at NTCIR - an annual workshop designed to enhance research into information-access technologies including cross-lingual information retrieval, text summarization etc. earlier this week.

The contest attracted the participation of universities, laboratories and research centers from five countries - Australia, India, Japan, Portugal and Vietnam - with the aim of building an open knowledge base with optimal artificial intelligence (AI) models to share with the technology community.

One of the institutes attending the event was Studio Ousia, a world-leading natural language AI in Japan famed for its LUKE model, its natural language processing ability surpassing models from Google and Facebook.

Participants had to classify 25 million articles in 30 different languages into 219 categories defined in Extended Named Entity, using categorized Japanese Wikipedia pages and interlanguage links to corresponding pages in target languages.

FPT.AI won the highest honors in three categories - English, Vietnamese, and Korean - in the Natural Language Processing (NLP) model competition for its advanced accuracy.

NLP technology focuses on studying language interactions between computers and humans, thereby offering solutions to increase human productivity, optimizing business operations.

The FPT.AI platform includes a solution ecosystem based on AI, featuring the multilingual chatbot system FPT.AI Conversation and virtual assistant of the Vietnamese voice-based interactive switchboard FPT.AI Virtual Agent For Call Center.